Abstract: The lack of remote sensing estimation models in vegetation coverage for desert regions of Xinjiang, China has brought great inconvenience to desertification monitoring and other related work. It is of positive and practical significance to study the empirical models for remote sensing estimation of vegetation coverage for promoting and improving ecological monitoring abilities in relevant areas. In this study, unmanned aerial vehicle (UAV) photography was firstly carried out in typical desert regions in Xinjiang, vegetation information (photosynthesis) was extracted by UAV images. Secondly, vegetation coverage was obtained at the pixel scale of high resolution satellite image by making the statistical units of vegetation coverage directly correspond to the pixels of high resolution satellite image in space. Lastly, based on the fitting relationship between vegetation coverage and the corresponding NDVI data derived from high resolution satellite images, the empirical models were established for estimating vegetation coverage of typical desert regions in Xinjiang. The deserts in the southern margin of the northern part of Fukang City and in the central plain of Karamay City were chosen as aerial shooting areas respectively. The relative flight height was set to 6-7 m and the ground resolution was 0.002-0.003 m. Satellite images which were shot on the same day and the day before UAV photographing, were selected and purchased for generating NDVI data. Agisoft Photoscan software was used for UAV image processing, ENVI and ArcGIS software packages were used for satellite remote sensing image processing and spatial analysis, Origin software was used for fitting analysis of data series and the vegetation information of UAV images was extracted by NDVI index. As a result, a linear model (y=149.86x-13.449,R²=0.735 3) was established for remote sensing estimation of vegetation coverage in northern sandy desert in Fukang based on NDVI derived from GF2 satellite image and a quadratic polynomial model （y=97.397x²+80.837x-5.210 9,R²=0.818） was established for remote sensing estimation of vegetation coverage in plain desert in Karamay based on NDVI derived from ZY1-02C satellite image. These models might provide necessary foundation and basis for monitoring and research of land use and land cover change, ecological environment change, desertification (or sandy desertification),vegetation mapping and other related work in Xinjiang. The combination of UAV images and satellite remotely sensed data and the spatial direct correspondence between vegetation coverage statistical unit and satellite pixel, could avoid the uncertainty caused by the process of representing vegetation coverage of satellite pixel with manual measurement data at different points in the previous related work. The proposed method made the spatial matching between vegetation coverage and satellite image pixels relatively more intuitive and accurate in the process of constructing remote sensing estimation model of vegetation coverage and it also provided an idea for similar work.

Key words: NDVI, desert, vegetation coverage, remote sensing estimation, empirical model